Asks read four rows per open pull, and home runs them twice #2780

Home now builds its ask candidates once per request, and the per-pull reads named in the issue are grouped. On the new test's seed, home's statement count stays flat as open pulls grow. Before the change it went from 101 to 143 statements between 4 and 16 open pulls.

What moved

• Grouped helpers (ids in, Map out, one query per chunk), each beside its singular read:

- activeRunsFor in src/core/runs.ts, beside activeRunFor. - handoversOf in src/crew/guard.ts, beside handoverOf. It runs one grouped count of crew change requests. Only a pull with three or more rounds goes on to the rest of the hand-over check, which is now handoverSince and is shared with handoverOf.

• askCandidatesFor (src/web/studio/asks.ts) uses both helpers instead of running activeRunFor and handoverOf once per pull.
• The app read: getAppByRepo ran once per card in decisionButtons. decisionsFor already reads the apps for every candidate's repository (appsForRepos), so it now sets a new preview field on Decision, and decisionButtons reads that field.
• Asks read once: the feed under "Needs you" (readFeed with withoutAsks) used to call askLinesFor and then drop every ask. candidatesFor (src/feed/items.ts) now takes a withAsks flag and skips the call. That leaves decisionsFor as home's only caller of askCandidatesFor.

Tests: test/asks-counts.test.ts checks four things:

• Home's statement count is the same for 4 open pulls as for 16. The seeded pulls are idle, crew-working, reviewed once, and on an autonomous repo.
• One GET / runs each grouped read exactly once.
• The grouped reads return the same results as the singular ones, including a pull that has been handed over.
• The preview link still appears once the repository becomes an app.

The first two fail on main. The type check passes, as do 39 related test files with 301 tests. Two tests in asks.test.ts ("a setting away from its default (#2029)") fail when run together with the other files, and fail the same way on main. They pass when asks.test.ts runs alone, so I did not change them.

Not done here: pulls that have a check result still cost about ten statements each. That covers the retry count on a failed check, the staleness and reviewer-wait checks on a green one, and the rest of the hand-over check, so the test seeds no such pulls. I opened #2781 for those. The repo row read the issue lists did not grow with pulls in what I measured.

To try it: bun test test/asks-counts.test.ts, or bun run bench/map-home.ts and compare the statement counts for GET / owner.

🤖 Generated with Claude Code

Closes #2772

Home now builds its ask candidates once per request, and the per-pull reads named in the issue are grouped. On the new test's seed, home's statement count stays flat as open pulls grow. Before the change it went from 101 to 143 statements between 4 and 16 open pulls.

What moved

• Grouped helpers (ids in, Map out, one query per chunk), each beside its singular read:

- activeRunsFor in src/core/runs.ts, beside activeRunFor. - handoversOf in src/crew/guard.ts, beside handoverOf. It runs one grouped count of crew change requests. Only a pull with three or more rounds goes on to the rest of the hand-over check, which is now handoverSince and is shared with handoverOf.

• askCandidatesFor (src/web/studio/asks.ts) uses both helpers instead of running activeRunFor and handoverOf once per pull.
• The app read: getAppByRepo ran once per card in decisionButtons. decisionsFor already reads the apps for every candidate's repository (appsForRepos), so it now sets a new preview field on Decision, and decisionButtons reads that field.
• Asks read once: the feed under "Needs you" (readFeed with withoutAsks) used to call askLinesFor and then drop every ask. candidatesFor (src/feed/items.ts) now takes a withAsks flag and skips the call. That leaves decisionsFor as home's only caller of askCandidatesFor.

Tests: test/asks-counts.test.ts checks four things:

• Home's statement count is the same for 4 open pulls as for 16. The seeded pulls are idle, crew-working, reviewed once, and on an autonomous repo.
• One GET / runs each grouped read exactly once.
• The grouped reads return the same results as the singular ones, including a pull that has been handed over.
• The preview link still appears once the repository becomes an app.

The first two fail on main. The type check passes, as do 39 related test files with 301 tests. Two tests in asks.test.ts ("a setting away from its default (#2029)") fail when run together with the other files, and fail the same way on main. They pass when asks.test.ts runs alone, so I did not change them.

Not done here: pulls that have a check result still cost about ten statements each. That covers the retry count on a failed check, the staleness and reviewer-wait checks on a green one, and the rest of the hand-over check, so the test seeds no such pulls. I opened #2781 for those. The repo row read the issue lists did not grow with pulls in what I measured.

To try it: bun test test/asks-counts.test.ts, or bun run bench/map-home.ts and compare the statement counts for GET / owner.

🤖 Generated with Claude Code